Anthropic can already retrain Mythos with "a fraction of the compute originally needed," says former researcher Coxon

Jacob Coxon, a former researcher at Anthropic and OpenAI, cited a slide from Anthropic's IPO roadshow showing superexponential algorithmic efficiency gains as a key reason for his departure. He highlighted that Anthropic's Mythos model can now be retrained with significantly less compute, signaling rapid progress in AI training efficiency. Coxon advocates for a pause in frontier AI development to allow competitors to catch up and prevent monopolization of human-level intelligence.